Open MahdiBM opened 2 months ago
By default the output is scaled with the scalingFactor to provide the per-real-iteration units - any output marked with *
signifies scaled output with the scaling factor. If you want to get the raw output, you need to provide the --scale
option;
--scale Show the metrics in the scale of the outer loop only (without applying the inner loop scalingFactor to the output)
The problem with raw output is that the numbers won't fit in the chart and will be truncated. That's the reason why I'm using scalingFactor in the first place.
Right, but that is a different issue?
I am responding to "What's wrong" - that output is scaled, so not sure what the problem is (truncation or output would be a different thing)?
Ok but then what the table says is still incorrect ... it shouldn't misinform by reporting the wrong amount of nanoseconds.
I'm also open to doing something to make the output fit the columns. Perhaps if there are trailing zeros, it should just use the e notation instead of truncating. Would that be reasonable?
See #276 for root.
The Benchmark:
The Result
✅ Pull request has no significant performance differences ✅
Benchmark check running at 2024-09-18 02:48:03 UTC
Baseline mmbm-no-twice-benchmark-run is EQUAL to the defined absolute baseline thresholds. (--check-absolute)
Baseline mmbm-no-twice-benchmark-run
MyBenchmarks
CountNumbersBenchmark
What's wrong?
Those
7710000
numbers are NOT in nanoseconds (1e-9), but in microseconds (1e-6). This is likely because i've set thescalingFactor
to.kilo
, aka1e3
. So that time label should have also been multiplied by the1e3
. Sonanos x kilos == micros
==1e-9 x 1e3 == 1e-6
.